The reliability of linear mixed models in genetic studies